Predicting Chromosome Conformation from Epigenetic Features
CSEF · 2015 Mathematics & Software
Overview
Objectives/Goals The objective is to predict the results of 40kb HinDIII Hi-C experimentation on the mouse ESC genome from 21 documented mouse epigenetic features. Methods/Materials Around 200 million 22-dimensional datapoints were drawn from the NIH histone database and from the Hi-C paper. Characteristics of the data required development of a novel machine learning meta-algorithm using inverse probability Metropolis-Hastings sampling, bagging with 100 independent samples, single-layer perceptrons optimized using Levenberg-Marquardt training, and intelligent weighting based on in-sample prediction correlation. Chromosome 1 was used for training, chromosome 2 for validation, and chromosomes 3 to X for testing. Results The model's predictions scored a Pearson correlation coefficient between 0.7 and 0.8 on all chromosomes. In-sample RMSE and out-of-sample RMSE decreased on all predictors over the 5 training epochs, and in-sample correlation and out-of-sample correlation increased on all predictors over the 5 training epochs. Conclusions/Discussion My method performs extremely well on this unusual dataset. In the next decade, spatial distance will be more insightful, cheaper and faster to compute, and easier to analyze in clinical situations than the DNA sequence itself.
Summary statement
A novel machine learning meta-algorithm is necessary and effective for studying the 3-dimensional conformation of the mammalian genome.
Help received
Used computing facilities at PSU Center for Comparative Genomics and Bioinformatics
Competition history
- CSEF 2015
Resources
Related projects
ISEF · 2018
A Novel Computational Model for Discovery of DNA Spatial Interactions: New Insights into Genomic Organization
ISEF · 2021
Refined Genome-wide Prediction of Transgenerational Epimutations and an Ensemble-based Solution to the Imbalanced Class Problem in Epigenetics
AJAS · 2019
Development of Semi-Supervised Machine Learning Models to Predict Enhancer Regions
CSEF · 2013
Identification of CpG Islands in a DNA Sequence Using a Hidden Markov Model Trained in MapReduce
ISEF · 2015
Computational Analysis of Neuronal Chromatin Structure and Nuclear PARP-1 and PAR Expression Provides Novel Marker for Detecting Learning Associated Changes in Mice
CSEF · 2012
HMM Model Usage to Determine Gene-Specific CpG Ratios and Location of Genes
ISEF · 2020
Activity-by-Contact Model to Predict Enhancer-Gene Connections: A Tool to Increase Our Understanding of Cancer
ISEF · 2026
Enhancing Gene Editing Safety: Leveraging 3D Genome Architecture and Epigenetics to Predict CRISPR-Cas9 Off-Target Mutations via Deep Learning
Closest projects by meaning, across every fair and year in the corpus.
Browse more like this
Source: California Science & Engineering Fair public projects